WO2021077725A1 - System and method for predicting motion state of surrounding vehicle based on driving intention - Google Patents

Abstract

A system and method for predicting the motion state of a surrounding vehicle based on a driving intention. The system comprises: feasible trajectory set generation, behavior intention generation, and predicted trajectory generation modules. The feasible trajectory set generation module is used for determining, according to the result of the global path planning, a current lane available for driving by a target vehicle to generate a feasible trajectory. A behavior intention inference module is used for predicting the probability that the target vehicle selects different lanes by analyzing the satisfaction of the target vehicle with different lanes, traffic regulations, and the state of the target vehicle. The predicted trajectory generation module is used for obtaining by fusion a predicted trajectory according to the generated feasible trajectory set and the result of the probability of a corresponding trajectory. The system solves the problem in the prior art of the omission of the interactive influence between the environment and the driver, and the dynamically changing environment when predicting the state of the vehicle.

Description

System and method for predicting motion state of surrounding vehicles based on driving intention Technical field

The invention belongs to the technical field of vehicle driving, and specifically refers to a system and method for predicting the motion state of surrounding vehicles based on driving intention.

Background technique

With the increasing number of cars, road traffic is gradually becoming denser and more complicated, which in turn leads to an increase in driving pressure, which reduces the ability of drivers to drive in normal traffic scenarios, and greatly increases the probability of traffic accidents. Lane-changing behavior is one of the important causes of traffic accidents and traffic congestion. Especially in urban areas, where the traffic density is high, lane-changing collision accidents are extremely likely to occur, and even serial rear-end collisions. The vast majority of lane-changing collision accidents are caused by inaccurate perception of the movement state and location information of the surrounding vehicles and making wrong driving decisions.

At present, intelligent vehicles can complete the lane changing process through advanced technology and avoid risks, which has become a key research direction to solve vehicle safety. However, the intelligent vehicle’s lane change decision process should not only consider the current state of its own vehicle and surrounding vehicles, but also obtain the final decision based on the prediction of the surrounding vehicle’s state in the future time domain. In terms of state prediction, the existing technology Most people think that in the prediction time domain, the surrounding vehicles are the process of maintaining the current behavior, and the other possible behaviors of the surrounding vehicles are not fully considered, thus ignoring the potential hazards.

Summary of the invention

In view of the above-mentioned shortcomings of the prior art, the purpose of the present invention is to provide a system and method for predicting the motion state of surrounding vehicles based on driving intentions of an autonomous vehicle, so as to solve the problem of neglecting the relationship between the environment and the driver when predicting the state of the vehicle in the prior art. The interactive influence and dynamic change of the environment.

In order to achieve the above objectives, the technical solutions adopted by the present invention are as follows:

The system for predicting the motion state of surrounding vehicles based on driving intention of the present invention includes: a feasible trajectory set generation module, a behavior intention inference module, and a predicted trajectory generation module;

The feasible trajectory set generation module, according to the result of the global path planning, determines the current lane of the target vehicle (that is, the predicted vehicle) to generate a feasible trajectory;

The behavior intention inference module predicts the probability of the target vehicle choosing different lanes by analyzing the target vehicle’s satisfaction with different lanes, traffic laws and the state of its own vehicle; because the target vehicle driver’s intention to change lanes is based on dynamic traffic Environment, not information at a certain point in time, historical information and current information can affect the output forecast results;

The predicted trajectory generating module merges the generated feasible trajectory set and the result of the probability of the corresponding trajectory to obtain the predicted trajectory.

Preferably, the feasible trajectory set generation module establishes a cost equation based on driving at a longitudinal speed and keeping it constant and entering a small steering angle to reach the lane center line of the desired lane, and the lateral kinematics model is a state space, so as to satisfy the cost The control input vector with the smallest equation value and the best feasible trajectory.

Preferably, the behavior intention inference module establishes a behavior intention inference model based on Recurrent Neural Network (RNN) and softmax regression analysis to obtain the probability of the corresponding trajectory in the above feasible trajectory set.

The method for predicting and controlling the motion state of surrounding vehicles based on driving intention inference according to the present invention, the steps are as follows:

1) Establish a cost equation based on driving at the longitudinal speed and keeping it constant and entering a small steering angle to reach the center line of the desired lane. The lateral kinematics model is the state space to solve the control input vector sum that satisfies the smallest value of the cost equation. The best feasible trajectory, a set of feasible trajectories can be generated according to all lanes;

2) Analyze the influencing factors of lane-changing intention based on the current state of the target vehicle's satisfaction with different lanes, combined with traffic laws and the state of the vehicle itself;

3) Define the output form of the intention inference model to represent left lane change, lane keeping, and right lane change respectively;

4) Establish the RNN intention inference model, take the factors analyzed in step 2) as the input of the model, and use the output form in step 3) as the output of the model to establish the calculation relationship of the model;

5) Use the data set {(x _t ,y _t )} ^{n to} train the network to obtain the weight coefficient matrix W, U, V and the bias vector b _h , b _{y in} step 4);

其中每个时刻的轨迹y _e,t,p可由

得到。 6): Based on the probability of different intentions obtained in steps 4) and 5) and the set of feasible trajectories obtained in step 1), the final predicted trajectory is obtained

The trajectory y _{e, t, p} at each moment can be

get.

其中，y _e为道路坐标系下的侧向位移，

分别为对应的侧向速度和侧向加速度，输入向量

表示侧向阶跃；T _s表示离散时间间隔，建立侧向运动的离散状态空间方程(1)如下： Preferably, the step 1) specifically includes: assuming that the longitudinal velocity remains unchanged, the selection state vector is

Among them, y _e is the lateral displacement in the road coordinate system,

Are the corresponding lateral velocity and lateral acceleration respectively, the input vector

Represents the lateral step; T _s represents the discrete time interval, and the discrete state space equation (1) of the lateral motion is established as follows:

Among them, k∈0,1,...,N-1 represents the discrete time step, and N represents the finite prediction time domain;

According to the input small steering angle to reach the lane center line of the desired lane, the cost equation (2) is given as follows:

Among them, Q≥0 and P≥0 respectively represent the process state and final state penalty factor, which is a positive semi-definite matrix, and R>0 is the input penalty factor, which is a positive definite matrix; χ _ref contains the information of the reference lane. According to the above, refer to The lateral velocity and acceleration should be 0;

Taking the current state of the vehicle as the initial state χ ₀ , the optimal control input sequence u ^* can be solved by the following formula (3):

Put u ^* into equation (1) to obtain the optimal state sequence χ ^* , and repeat the above steps according to different reference lanes to obtain a feasible trajectory set.

Preferably, the step 2) specifically includes:

21) Analyze the satisfaction degree of different lanes: the current lane satisfaction degree C _r,c ,C _p,c is given by the following formula:

Among them, x _e is the longitudinal position of the target vehicle, x _{p, c} , x _{r, c} are the longitudinal positions of the vehicle ahead and behind the current lane respectively, v _e , v _{r, c} are the longitudinal speeds of the target vehicle and the vehicle behind, respectively, L is the length of the vehicle body, and d _th is a preset value between the vehicle distances. If this value is exceeded, it is considered that there is no vehicle in front or behind in the lane;

Define the satisfaction levels of other adjacent lanes C _p,i ,C _r,i ,i∈{l,r}, where l represents the left lane, and r represents the right lane:

22) Analyze the impact of traffic laws on lane changing intentions, and consider the following factors:

The expected speed of the vehicle and the speed limit of the target lane, use C _v to express the driver’s satisfaction with the speed:

C _v =v _lim -v _desired (8)

Among them, v _lim represents the maximum speed of the target lane, and v _desired represents the desired speed of the current vehicle;

If the left and right lane lines are solid lines, lane changing is prohibited. C _{line is} used to indicate lane line information, solid indicates a solid line, and dashed indicates a dashed line:

C _line ∈{solid,dashed} (9);

23) Analyze the influence of the vehicle's own state on the intention to change lanes, and consider the following factors:

The current vehicle is related to the position of the centerline of the rightmost and leftmost lanes. If the driver is currently in the rightmost lane, the intention of changing lanes to the right will not occur. The feasibility of changing lanes is C _feasible . The distance from the centerline of the left lane to characterize:

C _feasible ＝y _e -y _road (10)

Among them, y _e represents the lateral position of the vehicle, and y _road represents the lateral position of the centerline of the leftmost lane;

来表示车辆状态的稳定性。 From the perspective of vehicle stability, if the vehicle's own state is unstable, it will not produce lane changing intentions, and use lateral acceleration

To indicate the stability of the vehicle state.

Optionally, the step 3) specifically includes: defining the output form of the intention inference model: based on the intention inference result of the lane, the result is coded in one-hot form, [1 0 0] means left lane change, [0 1 0 ] Means lane keeping, [0 0 1] means right lane change.

Preferably, the step 4) specifically includes: establishing an intention inference model based on RNN, and the influencing factors based on the analysis in step 2) are used as the input x _{t at} each time of the network:

The input of the input layer is a time series input X:

其中t时刻的隐状态h _t可由下式(11)计算得到： Given the input sequence, the hidden layer sequence

Among them, the hidden state h t at time _t can be calculated by the following formula (11):

h _t =tanh(Ux _t +Wh _t-1 +b _h ) (11)

Among them, U is the weight coefficient matrix between the input layer and the hidden layer, W is the weight coefficient of the cyclic connection in the _{hidden layer, and b h} is the bias vector of the hidden layer;

The output of the hidden layer is used as the input of the output layer, and the probability that the softmax layer will output different intention results

Wherein, V is the weight between the hidden layer and output layer weight coefficient matrix, b _y output layer as the offset vector.

Preferably, the specific training steps in step 5) are as follows:

Define the loss function between the true value and the predicted value as:

The weight coefficient matrix and bias vector can be obtained by solving the following equation (14):

The beneficial effects of the present invention:

The present invention considers the influence of other vehicles, roads and traffic laws on the future state of the vehicle when the intelligent vehicle is driving in the process of predicting the state of the surrounding vehicles, and considers the dynamic changes of the current driving environment, so as to fully and accurately understand the current driving traffic Information status, so as to make current decisions that are more in line with actual security.

Description of the drawings

Figure 1 is a block diagram of the principle of the system of the present invention.

Figure 2 is an example diagram of a set of feasible trajectories generated at a certain moment.

Fig. 3 is a calculation block diagram of the RNN network in the intention module of the present invention.

Detailed ways

In order to facilitate the understanding of those skilled in the art, the present invention will be further described below in conjunction with the embodiments and the drawings, and the content mentioned in the embodiments does not limit the present invention.

Referring to FIG. 1, a system for predicting the motion state of surrounding vehicles based on driving intention of the present invention includes: a feasible trajectory set generation module, a behavior intention inference module, and a predicted trajectory generation module;

The feasible trajectory set generation module, according to the result of the global path planning, determines the current lane of the target vehicle (that is, the predicted vehicle) to generate a feasible trajectory;

The behavior intention inference module predicts the probability of the target vehicle choosing different lanes by analyzing the target vehicle’s satisfaction with different lanes, traffic laws and the state of its own vehicle; because the target vehicle driver’s intention to change lanes is based on dynamic traffic Environment, not information at a certain point in time, historical information and current information can affect the output forecast results;

The predicted trajectory generation module merges the generated feasible trajectory set and the result of the probability of the corresponding trajectory to obtain the predicted trajectory.

Preferably, the feasible trajectory set generation module establishes a cost equation based on driving at a longitudinal speed and keeping it constant and entering a small steering angle to reach the lane center line of the desired lane, and the lateral kinematics model is a state space, so as to satisfy the cost The control input vector with the smallest equation value and the best feasible trajectory.

Preferably, the behavior intention inference module establishes a behavior intention inference model based on Recurrent Neural Network (RNN) and softmax regression analysis to obtain the probability of the corresponding trajectory in the above feasible trajectory set.

The method for predicting and controlling the motion state of surrounding vehicles based on driving intention inference of the present invention is based on the above system, and the steps are as follows:

1) Establish a cost equation based on driving at the longitudinal speed and keeping it constant and entering a small steering angle to reach the center line of the desired lane. The lateral kinematics model is the state space to solve the control input vector sum that satisfies the smallest value of the cost equation. The best feasible trajectory, a set of feasible trajectories can be generated according to all lanes;

其中，y _e为道路坐标系下的侧向位移，

分别为对应的侧向速度和侧向加速度，输入向量

表示侧向阶跃；T _s表示离散时间间隔，建立侧向运动的离散状态空间方程(1)如下： Referring to Figure 2, assuming that the longitudinal velocity remains constant, the selected state vector is

Among them, y _e is the lateral displacement in the road coordinate system,

Are the corresponding lateral velocity and lateral acceleration respectively, the input vector

Represents the lateral step; T _s represents the discrete time interval, and the discrete state space equation (1) of the lateral motion is established as follows:

Among them, k∈0,1,...,N-1 represents the discrete time step, and N represents the finite prediction time domain;

According to the input small steering angle to reach the lane center line of the desired lane, the cost equation (2) is given as follows:

Among them, Q≥0 and P≥0 respectively represent the process state and final state penalty factor, which is a positive semi-definite matrix, and R>0 is the input penalty factor, which is a positive definite matrix; χ _ref contains the information of the reference lane. According to the above, refer to The lateral velocity and acceleration should be 0;

Taking the current state of the vehicle as the initial state χ ₀ , the optimal control input sequence u ^* can be solved by the following formula (3):

Put u ^* into equation (1) to obtain the optimal state sequence χ ^* , and repeat the above steps according to different reference lanes to obtain a feasible trajectory set.

2) Analyze the influencing factors of lane changing intention based on the satisfaction of the target vehicle with different lanes under the current state, combined with the traffic laws and the state of the vehicle itself; specifically:

21) Analyze the satisfaction degree of different lanes: the current lane satisfaction degree C _r,c ,C _p,c is given by the following formula:

Among them, x _e is the longitudinal position of the target vehicle, x _{p, c} , x _{r, c} are the longitudinal positions of the vehicle ahead and behind the current lane respectively, v _e , v _{r, c} are the longitudinal speeds of the target vehicle and the vehicle behind, respectively, L is the length of the vehicle body, and d _th is a preset value between the vehicle distances. If this value is exceeded, it is considered that there is no vehicle in front or behind in the lane;

Define the satisfaction levels of other adjacent lanes C _p,i ,C _r,i ,i∈{l,r}, where l represents the left lane, and r represents the right lane:

22) Analyze the impact of traffic laws on lane changing intentions, and consider the following factors:

The expected speed of the vehicle and the speed limit of the target lane, use C _v to express the driver’s satisfaction with the speed:

C _v =v _lim -v _desired (8)

Among them, v _lim represents the maximum speed of the target lane, and v _desired represents the desired speed of the current vehicle;

If the left and right lane lines are solid lines, lane changing is prohibited. C _{line is} used to indicate lane line information, solid indicates a solid line, and dashed indicates a dashed line:

C _line ∈{solid,dashed} (9);

23) Analyze the influence of the vehicle's own state on the intention to change lanes, and consider the following factors:

The current vehicle is related to the position of the centerline of the rightmost and leftmost lanes. If the driver is currently in the rightmost lane, the intention of changing lanes to the right will not occur. The feasibility of changing lanes is C _feasible . The distance from the centerline of the left lane to characterize:

C _feasible ＝y _e -y _road (10)

Among them, y _e represents the lateral position of the vehicle, and y _road represents the lateral position of the centerline of the leftmost lane;

来表示车辆状态的稳定性。 From the perspective of vehicle stability, if the vehicle's own state is unstable, it will not produce lane changing intentions, and use lateral acceleration

To indicate the stability of the vehicle state.

3) Define the output form of the intention inference model to represent left lane change, lane keeping, and right lane change respectively;

Define the output form of the intention inference model: based on the intention inference result of the lane, encode the result in one-hot format, [1 0 0] means left lane change, [0 1 0] means lane keeping, [0 0 1] means Change lanes right

4) Establish the RNN intention inference model, take the factors analyzed in step 2) as the input of the model, and use the output form in step 3) as the output of the model to establish the calculation relationship of the model;

Referring to Figure 3, an intention inference model based on RNN is established, and the influencing factors based on the analysis in step 2) are used as the input x _{t at} each moment of the network:

The input of the input layer is a time series input X:

其中t时刻的隐状态h _t可由下式(11)计算得到： Given the input sequence, the hidden layer sequence

Among them, the hidden state h t at time _t can be calculated by the following formula (11):

h _t =tanh(Ux _t +Wh _t-1 +b _h ) (11)

Among them, U is the weight coefficient matrix between the input layer and the hidden layer, W is the weight coefficient of the cyclic connection in the _{hidden layer, and b h} is the bias vector of the hidden layer;

The output of the hidden layer is used as the input of the output layer, and the probability that the softmax layer will output different intention results

Wherein, V is the weight between the hidden layer and output layer weight coefficient matrix, b _y output layer as the offset vector.

5) Use the data set {(x _t ,y _t )} ^{n to} train the network to obtain the weight coefficient matrix W, U, V and the bias vector b _h , b _{y in} step 4);

Define the loss function between the true value and the predicted value as:

The weight coefficient matrix and bias vector can be obtained by solving the following equation (14):

其中每个时刻的轨迹y _e,t,p可由

得到。 6): Based on the probability of different intentions obtained in steps 4) and 5) and the set of feasible trajectories obtained in step 1), the final predicted trajectory is obtained

The trajectory y _{e, t, p} at each moment can be

get.

There are many specific applications of the present invention. The above are only the preferred embodiments of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements can be made. These Improvements should also be regarded as the protection scope of the present invention.

Claims (9)

A system for predicting the motion state of surrounding vehicles based on driving intention, which is characterized by comprising: a feasible trajectory set generation module, a behavior intention inference module, and a predicted trajectory generation module; The feasible trajectory set generation module, according to the result of the global path planning, determines the current lane of the target vehicle and generates a feasible trajectory; The behavioral intention inference module predicts the probability of the target vehicle choosing different lanes by analyzing the target vehicle’s satisfaction with different lanes, traffic laws and the state of its own vehicle; The predicted trajectory generation module fuses to obtain the predicted trajectory according to the result of the generated feasible trajectory set and the probability of the corresponding trajectory. The system for predicting the motion state of surrounding vehicles based on driving intent according to claim 1, wherein the feasible trajectory set generation module is based on the longitudinal speed and keeps constant and reaches the lane center of the desired lane by inputting a small steering angle. The cost equation is established by the line, and the lateral kinematics model is the state space to solve the control input vector that satisfies the smallest value of the cost equation and the best feasible trajectory. The surrounding vehicle motion state prediction system based on driving intention according to claim 1, wherein the behavior intention inference module establishes a behavior intention inference model based on cyclic neural network and softmax regression analysis, and obtains the above-mentioned feasible trajectory concentrated response The probability of the trajectory. A predictive control method for the motion state of surrounding vehicles based on driving intention inference, which is characterized in that the steps are as follows: 1) Establish a cost equation based on driving at the longitudinal speed and keeping it constant and entering a small steering angle to reach the center line of the desired lane. The lateral kinematics model is the state space to solve the control input vector sum that satisfies the smallest value of the cost equation. The best feasible trajectory, a set of feasible trajectories can be generated according to all lanes; 2) Analyze the influencing factors of lane-changing intention based on the current state of the target vehicle's satisfaction with different lanes, combined with traffic laws and the state of the vehicle itself; 3) Define the output form of the intention inference model to represent left lane change, lane keeping, and right lane change respectively; 4) Establish the RNN intention inference model, take the factors analyzed in step 2) as the input of the model, and use the output form in step 3) as the output of the model to establish the calculation relationship of the model; 5) Use the data set {(x _t ,y _t )} ^{n to} train the network to obtain the weight coefficient matrix W, U, V and the bias vector b _h , b _{y in} step 4); 6): Based on the probability of different intentions obtained in steps 4) and 5) and the set of feasible trajectories obtained in step 1), the final predicted trajectory is obtained

The trajectory y _{e, t, p} at each moment can be

get. The method for predicting and controlling the motion state of surrounding vehicles based on driving intention inference according to claim 4, wherein said step 1) specifically includes: assuming that the longitudinal speed remains unchanged, selecting the state vector as

Among them, y _e is the lateral displacement in the road coordinate system,

Are the corresponding lateral velocity and lateral acceleration respectively, the input vector

Represents the lateral step; T _s represents the discrete time interval, and the discrete state space equation (1) of the lateral motion is established as follows: χ _k+1 =Aχ _k +Bu _k (1)

Among them, k∈0,1,...,N-1 represents the discrete time step, and N represents the finite prediction time domain; According to the input small steering angle to reach the lane center line of the desired lane, the cost equation (2) is given as follows:

Among them, Q≥0 and P≥0 respectively represent the process state and final state penalty factor, which is a positive semi-definite matrix, and R>0 is the input penalty factor, which is a positive definite matrix; χ _ref contains the information of the reference lane. According to the above, refer to The lateral velocity and acceleration should be 0; Taking the current state of the vehicle as the initial state χ ₀ , the optimal control input sequence u ^* can be solved by the following formula (3):

Put u ^* into equation (1) to obtain the optimal state sequence χ ^* , and repeat the above steps according to different reference lanes to obtain a feasible trajectory set. The method for predicting and controlling the motion state of surrounding vehicles based on driving intention inference according to claim 4, wherein said step 2) specifically comprises: 21) Analyze the satisfaction degree of different lanes: the current lane satisfaction degree C _r,c ,C _p,c is given by the following formula:

Among them, x _e is the longitudinal position of the target vehicle, x _{p, c} , x _{r, c} are the longitudinal positions of the vehicle ahead and behind the current lane respectively, v _e , v _{r, c} are the longitudinal speeds of the target vehicle and the vehicle behind, respectively, L is the length of the vehicle body, and d _th is a preset value between the vehicle distances. If this value is exceeded, it is considered that there is no vehicle in front or behind in the lane; Define the satisfaction levels of other adjacent lanes C _p,i ,C _r,i ,i∈{l,r}, where l represents the left lane, and r represents the right lane:

22) Analyze the impact of traffic laws on lane changing intentions, and consider the following factors: The expected speed of the vehicle and the speed limit of the target lane, use C _v to express the driver’s satisfaction with the speed: C _v =v _lim -v _desired (8) Among them, v _lim represents the highest vehicle speed in the target lane, and v _desired represents the desired speed of the current vehicle; If the left and right lane lines are solid lines, lane changing is prohibited. C _{line is} used to indicate lane line information, solid indicates a solid line, and dashed indicates a dashed line: C _line ∈{solid,dashed} (9); 23) Analyze the influence of the vehicle's own state on the intention to change lanes, and consider the following factors: The current vehicle is related to the position of the centerline of the rightmost and leftmost lanes. If the driver is currently in the rightmost lane, the intention of changing lanes to the right will not be generated. The feasibility of changing lanes is C _feasible . The distance from the centerline of the left lane to characterize: C _feasible ＝y _e -y _road (10) Among them, y _e represents the lateral position of the vehicle, and y _road represents the lateral position of the center line of the leftmost lane; From the perspective of vehicle stability, if the vehicle's own state is unstable, it will not produce lane changing intentions, and use lateral acceleration

To indicate the stability of the vehicle state. The method for predicting and controlling the motion state of surrounding vehicles based on driving intention inference according to claim 4, wherein the step 3) specifically includes: defining the output form of the intention inference model: the intention inference result based on the lane is used, and the result is used One-hot format code, [1 0 0] means left lane change, [0 1 0] means lane keeping, [0 0 1] means right lane change. The method for predicting and controlling the motion state of surrounding vehicles based on driving intention inference according to claim 4, wherein said step 4) specifically comprises: establishing an intention inference model based on RNN, and based on the influencing factors analyzed in step 2) _{As the input x t at} each moment of the network:

The input of the input layer is a time series input X:

Given the input sequence, the hidden layer sequence

Among them, the hidden state h t at time _t can be calculated by the following formula (11): h _t =tanh(Ux _t +Wh _t-1 +b _h ) (11) Among them, U is the weight coefficient matrix between the input layer and the hidden layer, W is the weight coefficient of the cyclic connection in the _{hidden layer, and b h} is the bias vector of the hidden layer; The output of the hidden layer is used as the input of the output layer, and the probability that the softmax layer will output different intention results

Wherein, V is the weight between the hidden layer and output layer weight coefficient matrix, b _y output layer as the offset vector. The method for predicting and controlling the motion state of surrounding vehicles based on driving intention inference according to claim 4, wherein the specific training steps in step 5) are as follows: Define the loss function between the true value and the predicted value as:

The weight coefficient matrix and bias vector can be obtained by solving the following equation (14):

Images